1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34#include <linux/slab.h>
35#include <linux/delay.h>
36#include <linux/blkdev.h>
37#include <linux/module.h>
38#include <linux/seq_file.h>
39#include <linux/ratelimit.h>
40#include "md.h"
41#include "raid1.h"
42#include "bitmap.h"
43
44
45
46
47#define NR_RAID1_BIOS 256
48
49
50
51
52
53
54#define IO_BLOCKED ((struct bio *)1)
55
56
57
58
59#define IO_MADE_GOOD ((struct bio *)2)
60
61#define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
62
63
64
65
66
67static int max_queued_requests = 1024;
68
69static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
70 sector_t bi_sector);
71static void lower_barrier(struct r1conf *conf);
72
73static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
74{
75 struct pool_info *pi = data;
76 int size = offsetof(struct r1bio, bios[pi->raid_disks]);
77
78
79 return kzalloc(size, gfp_flags);
80}
81
82static void r1bio_pool_free(void *r1_bio, void *data)
83{
84 kfree(r1_bio);
85}
86
87#define RESYNC_BLOCK_SIZE (64*1024)
88#define RESYNC_DEPTH 32
89#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
90#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
91#define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)
92#define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)
93#define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW)
94#define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9)
95#define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS)
96
97static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
98{
99 struct pool_info *pi = data;
100 struct r1bio *r1_bio;
101 struct bio *bio;
102 int need_pages;
103 int i, j;
104
105 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
106 if (!r1_bio)
107 return NULL;
108
109
110
111
112 for (j = pi->raid_disks ; j-- ; ) {
113 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
114 if (!bio)
115 goto out_free_bio;
116 r1_bio->bios[j] = bio;
117 }
118
119
120
121
122
123
124 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
125 need_pages = pi->raid_disks;
126 else
127 need_pages = 1;
128 for (j = 0; j < need_pages; j++) {
129 bio = r1_bio->bios[j];
130 bio->bi_vcnt = RESYNC_PAGES;
131
132 if (bio_alloc_pages(bio, gfp_flags))
133 goto out_free_pages;
134 }
135
136 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
137 for (i=0; i<RESYNC_PAGES ; i++)
138 for (j=1; j<pi->raid_disks; j++)
139 r1_bio->bios[j]->bi_io_vec[i].bv_page =
140 r1_bio->bios[0]->bi_io_vec[i].bv_page;
141 }
142
143 r1_bio->master_bio = NULL;
144
145 return r1_bio;
146
147out_free_pages:
148 while (--j >= 0) {
149 struct bio_vec *bv;
150
151 bio_for_each_segment_all(bv, r1_bio->bios[j], i)
152 __free_page(bv->bv_page);
153 }
154
155out_free_bio:
156 while (++j < pi->raid_disks)
157 bio_put(r1_bio->bios[j]);
158 r1bio_pool_free(r1_bio, data);
159 return NULL;
160}
161
162static void r1buf_pool_free(void *__r1_bio, void *data)
163{
164 struct pool_info *pi = data;
165 int i,j;
166 struct r1bio *r1bio = __r1_bio;
167
168 for (i = 0; i < RESYNC_PAGES; i++)
169 for (j = pi->raid_disks; j-- ;) {
170 if (j == 0 ||
171 r1bio->bios[j]->bi_io_vec[i].bv_page !=
172 r1bio->bios[0]->bi_io_vec[i].bv_page)
173 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
174 }
175 for (i=0 ; i < pi->raid_disks; i++)
176 bio_put(r1bio->bios[i]);
177
178 r1bio_pool_free(r1bio, data);
179}
180
181static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
182{
183 int i;
184
185 for (i = 0; i < conf->raid_disks * 2; i++) {
186 struct bio **bio = r1_bio->bios + i;
187 if (!BIO_SPECIAL(*bio))
188 bio_put(*bio);
189 *bio = NULL;
190 }
191}
192
193static void free_r1bio(struct r1bio *r1_bio)
194{
195 struct r1conf *conf = r1_bio->mddev->private;
196
197 put_all_bios(conf, r1_bio);
198 mempool_free(r1_bio, conf->r1bio_pool);
199}
200
201static void put_buf(struct r1bio *r1_bio)
202{
203 struct r1conf *conf = r1_bio->mddev->private;
204 int i;
205
206 for (i = 0; i < conf->raid_disks * 2; i++) {
207 struct bio *bio = r1_bio->bios[i];
208 if (bio->bi_end_io)
209 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
210 }
211
212 mempool_free(r1_bio, conf->r1buf_pool);
213
214 lower_barrier(conf);
215}
216
217static void reschedule_retry(struct r1bio *r1_bio)
218{
219 unsigned long flags;
220 struct mddev *mddev = r1_bio->mddev;
221 struct r1conf *conf = mddev->private;
222
223 spin_lock_irqsave(&conf->device_lock, flags);
224 list_add(&r1_bio->retry_list, &conf->retry_list);
225 conf->nr_queued ++;
226 spin_unlock_irqrestore(&conf->device_lock, flags);
227
228 wake_up(&conf->wait_barrier);
229 md_wakeup_thread(mddev->thread);
230}
231
232
233
234
235
236
237static void call_bio_endio(struct r1bio *r1_bio)
238{
239 struct bio *bio = r1_bio->master_bio;
240 int done;
241 struct r1conf *conf = r1_bio->mddev->private;
242 sector_t start_next_window = r1_bio->start_next_window;
243 sector_t bi_sector = bio->bi_iter.bi_sector;
244
245 if (bio->bi_phys_segments) {
246 unsigned long flags;
247 spin_lock_irqsave(&conf->device_lock, flags);
248 bio->bi_phys_segments--;
249 done = (bio->bi_phys_segments == 0);
250 spin_unlock_irqrestore(&conf->device_lock, flags);
251
252
253
254
255 wake_up(&conf->wait_barrier);
256 } else
257 done = 1;
258
259 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
260 bio->bi_error = -EIO;
261
262 if (done) {
263 bio_endio(bio);
264
265
266
267
268 allow_barrier(conf, start_next_window, bi_sector);
269 }
270}
271
272static void raid_end_bio_io(struct r1bio *r1_bio)
273{
274 struct bio *bio = r1_bio->master_bio;
275
276
277 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
278 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
279 (bio_data_dir(bio) == WRITE) ? "write" : "read",
280 (unsigned long long) bio->bi_iter.bi_sector,
281 (unsigned long long) bio_end_sector(bio) - 1);
282
283 call_bio_endio(r1_bio);
284 }
285 free_r1bio(r1_bio);
286}
287
288
289
290
291static inline void update_head_pos(int disk, struct r1bio *r1_bio)
292{
293 struct r1conf *conf = r1_bio->mddev->private;
294
295 conf->mirrors[disk].head_position =
296 r1_bio->sector + (r1_bio->sectors);
297}
298
299
300
301
302static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
303{
304 int mirror;
305 struct r1conf *conf = r1_bio->mddev->private;
306 int raid_disks = conf->raid_disks;
307
308 for (mirror = 0; mirror < raid_disks * 2; mirror++)
309 if (r1_bio->bios[mirror] == bio)
310 break;
311
312 BUG_ON(mirror == raid_disks * 2);
313 update_head_pos(mirror, r1_bio);
314
315 return mirror;
316}
317
318static void raid1_end_read_request(struct bio *bio)
319{
320 int uptodate = !bio->bi_error;
321 struct r1bio *r1_bio = bio->bi_private;
322 int mirror;
323 struct r1conf *conf = r1_bio->mddev->private;
324
325 mirror = r1_bio->read_disk;
326
327
328
329 update_head_pos(mirror, r1_bio);
330
331 if (uptodate)
332 set_bit(R1BIO_Uptodate, &r1_bio->state);
333 else {
334
335
336
337
338 unsigned long flags;
339 spin_lock_irqsave(&conf->device_lock, flags);
340 if (r1_bio->mddev->degraded == conf->raid_disks ||
341 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
342 test_bit(In_sync, &conf->mirrors[mirror].rdev->flags)))
343 uptodate = 1;
344 spin_unlock_irqrestore(&conf->device_lock, flags);
345 }
346
347 if (uptodate) {
348 raid_end_bio_io(r1_bio);
349 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
350 } else {
351
352
353
354 char b[BDEVNAME_SIZE];
355 printk_ratelimited(
356 KERN_ERR "md/raid1:%s: %s: "
357 "rescheduling sector %llu\n",
358 mdname(conf->mddev),
359 bdevname(conf->mirrors[mirror].rdev->bdev,
360 b),
361 (unsigned long long)r1_bio->sector);
362 set_bit(R1BIO_ReadError, &r1_bio->state);
363 reschedule_retry(r1_bio);
364
365 }
366}
367
368static void close_write(struct r1bio *r1_bio)
369{
370
371 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
372
373 int i = r1_bio->behind_page_count;
374 while (i--)
375 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
376 kfree(r1_bio->behind_bvecs);
377 r1_bio->behind_bvecs = NULL;
378 }
379
380 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
381 r1_bio->sectors,
382 !test_bit(R1BIO_Degraded, &r1_bio->state),
383 test_bit(R1BIO_BehindIO, &r1_bio->state));
384 md_write_end(r1_bio->mddev);
385}
386
387static void r1_bio_write_done(struct r1bio *r1_bio)
388{
389 if (!atomic_dec_and_test(&r1_bio->remaining))
390 return;
391
392 if (test_bit(R1BIO_WriteError, &r1_bio->state))
393 reschedule_retry(r1_bio);
394 else {
395 close_write(r1_bio);
396 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
397 reschedule_retry(r1_bio);
398 else
399 raid_end_bio_io(r1_bio);
400 }
401}
402
403static void raid1_end_write_request(struct bio *bio)
404{
405 struct r1bio *r1_bio = bio->bi_private;
406 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
407 struct r1conf *conf = r1_bio->mddev->private;
408 struct bio *to_put = NULL;
409
410 mirror = find_bio_disk(r1_bio, bio);
411
412
413
414
415 if (bio->bi_error) {
416 set_bit(WriteErrorSeen,
417 &conf->mirrors[mirror].rdev->flags);
418 if (!test_and_set_bit(WantReplacement,
419 &conf->mirrors[mirror].rdev->flags))
420 set_bit(MD_RECOVERY_NEEDED, &
421 conf->mddev->recovery);
422
423 set_bit(R1BIO_WriteError, &r1_bio->state);
424 } else {
425
426
427
428
429
430
431
432
433
434
435 sector_t first_bad;
436 int bad_sectors;
437
438 r1_bio->bios[mirror] = NULL;
439 to_put = bio;
440
441
442
443
444
445
446
447
448 if (test_bit(In_sync, &conf->mirrors[mirror].rdev->flags) &&
449 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))
450 set_bit(R1BIO_Uptodate, &r1_bio->state);
451
452
453 if (is_badblock(conf->mirrors[mirror].rdev,
454 r1_bio->sector, r1_bio->sectors,
455 &first_bad, &bad_sectors)) {
456 r1_bio->bios[mirror] = IO_MADE_GOOD;
457 set_bit(R1BIO_MadeGood, &r1_bio->state);
458 }
459 }
460
461 if (behind) {
462 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
463 atomic_dec(&r1_bio->behind_remaining);
464
465
466
467
468
469
470
471
472 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
473 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
474
475 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
476 struct bio *mbio = r1_bio->master_bio;
477 pr_debug("raid1: behind end write sectors"
478 " %llu-%llu\n",
479 (unsigned long long) mbio->bi_iter.bi_sector,
480 (unsigned long long) bio_end_sector(mbio) - 1);
481 call_bio_endio(r1_bio);
482 }
483 }
484 }
485 if (r1_bio->bios[mirror] == NULL)
486 rdev_dec_pending(conf->mirrors[mirror].rdev,
487 conf->mddev);
488
489
490
491
492
493 r1_bio_write_done(r1_bio);
494
495 if (to_put)
496 bio_put(to_put);
497}
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
514{
515 const sector_t this_sector = r1_bio->sector;
516 int sectors;
517 int best_good_sectors;
518 int best_disk, best_dist_disk, best_pending_disk;
519 int has_nonrot_disk;
520 int disk;
521 sector_t best_dist;
522 unsigned int min_pending;
523 struct md_rdev *rdev;
524 int choose_first;
525 int choose_next_idle;
526
527 rcu_read_lock();
528
529
530
531
532
533 retry:
534 sectors = r1_bio->sectors;
535 best_disk = -1;
536 best_dist_disk = -1;
537 best_dist = MaxSector;
538 best_pending_disk = -1;
539 min_pending = UINT_MAX;
540 best_good_sectors = 0;
541 has_nonrot_disk = 0;
542 choose_next_idle = 0;
543
544 if ((conf->mddev->recovery_cp < this_sector + sectors) ||
545 (mddev_is_clustered(conf->mddev) &&
546 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector,
547 this_sector + sectors)))
548 choose_first = 1;
549 else
550 choose_first = 0;
551
552 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
553 sector_t dist;
554 sector_t first_bad;
555 int bad_sectors;
556 unsigned int pending;
557 bool nonrot;
558
559 rdev = rcu_dereference(conf->mirrors[disk].rdev);
560 if (r1_bio->bios[disk] == IO_BLOCKED
561 || rdev == NULL
562 || test_bit(Faulty, &rdev->flags))
563 continue;
564 if (!test_bit(In_sync, &rdev->flags) &&
565 rdev->recovery_offset < this_sector + sectors)
566 continue;
567 if (test_bit(WriteMostly, &rdev->flags)) {
568
569
570 if (best_dist_disk < 0) {
571 if (is_badblock(rdev, this_sector, sectors,
572 &first_bad, &bad_sectors)) {
573 if (first_bad < this_sector)
574
575 continue;
576 best_good_sectors = first_bad - this_sector;
577 } else
578 best_good_sectors = sectors;
579 best_dist_disk = disk;
580 best_pending_disk = disk;
581 }
582 continue;
583 }
584
585
586
587 if (is_badblock(rdev, this_sector, sectors,
588 &first_bad, &bad_sectors)) {
589 if (best_dist < MaxSector)
590
591 continue;
592 if (first_bad <= this_sector) {
593
594
595
596
597 bad_sectors -= (this_sector - first_bad);
598 if (choose_first && sectors > bad_sectors)
599 sectors = bad_sectors;
600 if (best_good_sectors > sectors)
601 best_good_sectors = sectors;
602
603 } else {
604 sector_t good_sectors = first_bad - this_sector;
605 if (good_sectors > best_good_sectors) {
606 best_good_sectors = good_sectors;
607 best_disk = disk;
608 }
609 if (choose_first)
610 break;
611 }
612 continue;
613 } else
614 best_good_sectors = sectors;
615
616 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
617 has_nonrot_disk |= nonrot;
618 pending = atomic_read(&rdev->nr_pending);
619 dist = abs(this_sector - conf->mirrors[disk].head_position);
620 if (choose_first) {
621 best_disk = disk;
622 break;
623 }
624
625 if (conf->mirrors[disk].next_seq_sect == this_sector
626 || dist == 0) {
627 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
628 struct raid1_info *mirror = &conf->mirrors[disk];
629
630 best_disk = disk;
631
632
633
634
635
636
637
638
639
640
641
642
643
644 if (nonrot && opt_iosize > 0 &&
645 mirror->seq_start != MaxSector &&
646 mirror->next_seq_sect > opt_iosize &&
647 mirror->next_seq_sect - opt_iosize >=
648 mirror->seq_start) {
649 choose_next_idle = 1;
650 continue;
651 }
652 break;
653 }
654
655 if (pending == 0) {
656 best_disk = disk;
657 break;
658 }
659
660 if (choose_next_idle)
661 continue;
662
663 if (min_pending > pending) {
664 min_pending = pending;
665 best_pending_disk = disk;
666 }
667
668 if (dist < best_dist) {
669 best_dist = dist;
670 best_dist_disk = disk;
671 }
672 }
673
674
675
676
677
678
679
680 if (best_disk == -1) {
681 if (has_nonrot_disk)
682 best_disk = best_pending_disk;
683 else
684 best_disk = best_dist_disk;
685 }
686
687 if (best_disk >= 0) {
688 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
689 if (!rdev)
690 goto retry;
691 atomic_inc(&rdev->nr_pending);
692 if (test_bit(Faulty, &rdev->flags)) {
693
694
695
696 rdev_dec_pending(rdev, conf->mddev);
697 goto retry;
698 }
699 sectors = best_good_sectors;
700
701 if (conf->mirrors[best_disk].next_seq_sect != this_sector)
702 conf->mirrors[best_disk].seq_start = this_sector;
703
704 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
705 }
706 rcu_read_unlock();
707 *max_sectors = sectors;
708
709 return best_disk;
710}
711
712static int raid1_congested(struct mddev *mddev, int bits)
713{
714 struct r1conf *conf = mddev->private;
715 int i, ret = 0;
716
717 if ((bits & (1 << WB_async_congested)) &&
718 conf->pending_count >= max_queued_requests)
719 return 1;
720
721 rcu_read_lock();
722 for (i = 0; i < conf->raid_disks * 2; i++) {
723 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
724 if (rdev && !test_bit(Faulty, &rdev->flags)) {
725 struct request_queue *q = bdev_get_queue(rdev->bdev);
726
727 BUG_ON(!q);
728
729
730
731
732 if ((bits & (1 << WB_async_congested)) || 1)
733 ret |= bdi_congested(&q->backing_dev_info, bits);
734 else
735 ret &= bdi_congested(&q->backing_dev_info, bits);
736 }
737 }
738 rcu_read_unlock();
739 return ret;
740}
741
742static void flush_pending_writes(struct r1conf *conf)
743{
744
745
746
747 spin_lock_irq(&conf->device_lock);
748
749 if (conf->pending_bio_list.head) {
750 struct bio *bio;
751 bio = bio_list_get(&conf->pending_bio_list);
752 conf->pending_count = 0;
753 spin_unlock_irq(&conf->device_lock);
754
755
756 bitmap_unplug(conf->mddev->bitmap);
757 wake_up(&conf->wait_barrier);
758
759 while (bio) {
760 struct bio *next = bio->bi_next;
761 bio->bi_next = NULL;
762 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
763 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
764
765 bio_endio(bio);
766 else
767 generic_make_request(bio);
768 bio = next;
769 }
770 } else
771 spin_unlock_irq(&conf->device_lock);
772}
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
796{
797 spin_lock_irq(&conf->resync_lock);
798
799
800 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
801 conf->resync_lock);
802
803
804 conf->barrier++;
805 conf->next_resync = sector_nr;
806
807
808
809
810
811
812
813
814
815
816 wait_event_lock_irq(conf->wait_barrier,
817 !conf->array_frozen &&
818 conf->barrier < RESYNC_DEPTH &&
819 conf->current_window_requests == 0 &&
820 (conf->start_next_window >=
821 conf->next_resync + RESYNC_SECTORS),
822 conf->resync_lock);
823
824 conf->nr_pending++;
825 spin_unlock_irq(&conf->resync_lock);
826}
827
828static void lower_barrier(struct r1conf *conf)
829{
830 unsigned long flags;
831 BUG_ON(conf->barrier <= 0);
832 spin_lock_irqsave(&conf->resync_lock, flags);
833 conf->barrier--;
834 conf->nr_pending--;
835 spin_unlock_irqrestore(&conf->resync_lock, flags);
836 wake_up(&conf->wait_barrier);
837}
838
839static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio)
840{
841 bool wait = false;
842
843 if (conf->array_frozen || !bio)
844 wait = true;
845 else if (conf->barrier && bio_data_dir(bio) == WRITE) {
846 if ((conf->mddev->curr_resync_completed
847 >= bio_end_sector(bio)) ||
848 (conf->next_resync + NEXT_NORMALIO_DISTANCE
849 <= bio->bi_iter.bi_sector))
850 wait = false;
851 else
852 wait = true;
853 }
854
855 return wait;
856}
857
858static sector_t wait_barrier(struct r1conf *conf, struct bio *bio)
859{
860 sector_t sector = 0;
861
862 spin_lock_irq(&conf->resync_lock);
863 if (need_to_wait_for_sync(conf, bio)) {
864 conf->nr_waiting++;
865
866
867
868
869
870
871
872
873
874 wait_event_lock_irq(conf->wait_barrier,
875 !conf->array_frozen &&
876 (!conf->barrier ||
877 ((conf->start_next_window <
878 conf->next_resync + RESYNC_SECTORS) &&
879 current->bio_list &&
880 !bio_list_empty(current->bio_list))),
881 conf->resync_lock);
882 conf->nr_waiting--;
883 }
884
885 if (bio && bio_data_dir(bio) == WRITE) {
886 if (bio->bi_iter.bi_sector >= conf->next_resync) {
887 if (conf->start_next_window == MaxSector)
888 conf->start_next_window =
889 conf->next_resync +
890 NEXT_NORMALIO_DISTANCE;
891
892 if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE)
893 <= bio->bi_iter.bi_sector)
894 conf->next_window_requests++;
895 else
896 conf->current_window_requests++;
897 sector = conf->start_next_window;
898 }
899 }
900
901 conf->nr_pending++;
902 spin_unlock_irq(&conf->resync_lock);
903 return sector;
904}
905
906static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
907 sector_t bi_sector)
908{
909 unsigned long flags;
910
911 spin_lock_irqsave(&conf->resync_lock, flags);
912 conf->nr_pending--;
913 if (start_next_window) {
914 if (start_next_window == conf->start_next_window) {
915 if (conf->start_next_window + NEXT_NORMALIO_DISTANCE
916 <= bi_sector)
917 conf->next_window_requests--;
918 else
919 conf->current_window_requests--;
920 } else
921 conf->current_window_requests--;
922
923 if (!conf->current_window_requests) {
924 if (conf->next_window_requests) {
925 conf->current_window_requests =
926 conf->next_window_requests;
927 conf->next_window_requests = 0;
928 conf->start_next_window +=
929 NEXT_NORMALIO_DISTANCE;
930 } else
931 conf->start_next_window = MaxSector;
932 }
933 }
934 spin_unlock_irqrestore(&conf->resync_lock, flags);
935 wake_up(&conf->wait_barrier);
936}
937
938static void freeze_array(struct r1conf *conf, int extra)
939{
940
941
942
943
944
945
946
947
948
949
950
951 spin_lock_irq(&conf->resync_lock);
952 conf->array_frozen = 1;
953 wait_event_lock_irq_cmd(conf->wait_barrier,
954 conf->nr_pending == conf->nr_queued+extra,
955 conf->resync_lock,
956 flush_pending_writes(conf));
957 spin_unlock_irq(&conf->resync_lock);
958}
959static void unfreeze_array(struct r1conf *conf)
960{
961
962 spin_lock_irq(&conf->resync_lock);
963 conf->array_frozen = 0;
964 wake_up(&conf->wait_barrier);
965 spin_unlock_irq(&conf->resync_lock);
966}
967
968
969
970static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
971{
972 int i;
973 struct bio_vec *bvec;
974 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
975 GFP_NOIO);
976 if (unlikely(!bvecs))
977 return;
978
979 bio_for_each_segment_all(bvec, bio, i) {
980 bvecs[i] = *bvec;
981 bvecs[i].bv_page = alloc_page(GFP_NOIO);
982 if (unlikely(!bvecs[i].bv_page))
983 goto do_sync_io;
984 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
985 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
986 kunmap(bvecs[i].bv_page);
987 kunmap(bvec->bv_page);
988 }
989 r1_bio->behind_bvecs = bvecs;
990 r1_bio->behind_page_count = bio->bi_vcnt;
991 set_bit(R1BIO_BehindIO, &r1_bio->state);
992 return;
993
994do_sync_io:
995 for (i = 0; i < bio->bi_vcnt; i++)
996 if (bvecs[i].bv_page)
997 put_page(bvecs[i].bv_page);
998 kfree(bvecs);
999 pr_debug("%dB behind alloc failed, doing sync I/O\n",
1000 bio->bi_iter.bi_size);
1001}
1002
1003struct raid1_plug_cb {
1004 struct blk_plug_cb cb;
1005 struct bio_list pending;
1006 int pending_cnt;
1007};
1008
1009static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
1010{
1011 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
1012 cb);
1013 struct mddev *mddev = plug->cb.data;
1014 struct r1conf *conf = mddev->private;
1015 struct bio *bio;
1016
1017 if (from_schedule || current->bio_list) {
1018 spin_lock_irq(&conf->device_lock);
1019 bio_list_merge(&conf->pending_bio_list, &plug->pending);
1020 conf->pending_count += plug->pending_cnt;
1021 spin_unlock_irq(&conf->device_lock);
1022 wake_up(&conf->wait_barrier);
1023 md_wakeup_thread(mddev->thread);
1024 kfree(plug);
1025 return;
1026 }
1027
1028
1029 bio = bio_list_get(&plug->pending);
1030 bitmap_unplug(mddev->bitmap);
1031 wake_up(&conf->wait_barrier);
1032
1033 while (bio) {
1034 struct bio *next = bio->bi_next;
1035 bio->bi_next = NULL;
1036 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
1037 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
1038
1039 bio_endio(bio);
1040 else
1041 generic_make_request(bio);
1042 bio = next;
1043 }
1044 kfree(plug);
1045}
1046
1047static void raid1_make_request(struct mddev *mddev, struct bio * bio)
1048{
1049 struct r1conf *conf = mddev->private;
1050 struct raid1_info *mirror;
1051 struct r1bio *r1_bio;
1052 struct bio *read_bio;
1053 int i, disks;
1054 struct bitmap *bitmap;
1055 unsigned long flags;
1056 const int rw = bio_data_dir(bio);
1057 const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
1058 const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA));
1059 const unsigned long do_discard = (bio->bi_rw
1060 & (REQ_DISCARD | REQ_SECURE));
1061 const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME);
1062 struct md_rdev *blocked_rdev;
1063 struct blk_plug_cb *cb;
1064 struct raid1_plug_cb *plug = NULL;
1065 int first_clone;
1066 int sectors_handled;
1067 int max_sectors;
1068 sector_t start_next_window;
1069
1070
1071
1072
1073
1074
1075
1076 md_write_start(mddev, bio);
1077
1078 if (bio_data_dir(bio) == WRITE &&
1079 ((bio_end_sector(bio) > mddev->suspend_lo &&
1080 bio->bi_iter.bi_sector < mddev->suspend_hi) ||
1081 (mddev_is_clustered(mddev) &&
1082 md_cluster_ops->area_resyncing(mddev, WRITE,
1083 bio->bi_iter.bi_sector, bio_end_sector(bio))))) {
1084
1085
1086
1087
1088 DEFINE_WAIT(w);
1089 for (;;) {
1090 flush_signals(current);
1091 prepare_to_wait(&conf->wait_barrier,
1092 &w, TASK_INTERRUPTIBLE);
1093 if (bio_end_sector(bio) <= mddev->suspend_lo ||
1094 bio->bi_iter.bi_sector >= mddev->suspend_hi ||
1095 (mddev_is_clustered(mddev) &&
1096 !md_cluster_ops->area_resyncing(mddev, WRITE,
1097 bio->bi_iter.bi_sector, bio_end_sector(bio))))
1098 break;
1099 schedule();
1100 }
1101 finish_wait(&conf->wait_barrier, &w);
1102 }
1103
1104 start_next_window = wait_barrier(conf, bio);
1105
1106 bitmap = mddev->bitmap;
1107
1108
1109
1110
1111
1112
1113 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1114
1115 r1_bio->master_bio = bio;
1116 r1_bio->sectors = bio_sectors(bio);
1117 r1_bio->state = 0;
1118 r1_bio->mddev = mddev;
1119 r1_bio->sector = bio->bi_iter.bi_sector;
1120
1121
1122
1123
1124
1125
1126
1127
1128 bio->bi_phys_segments = 0;
1129 bio_clear_flag(bio, BIO_SEG_VALID);
1130
1131 if (rw == READ) {
1132
1133
1134
1135 int rdisk;
1136
1137read_again:
1138 rdisk = read_balance(conf, r1_bio, &max_sectors);
1139
1140 if (rdisk < 0) {
1141
1142 raid_end_bio_io(r1_bio);
1143 return;
1144 }
1145 mirror = conf->mirrors + rdisk;
1146
1147 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1148 bitmap) {
1149
1150
1151
1152
1153 wait_event(bitmap->behind_wait,
1154 atomic_read(&bitmap->behind_writes) == 0);
1155 }
1156 r1_bio->read_disk = rdisk;
1157 r1_bio->start_next_window = 0;
1158
1159 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1160 bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector,
1161 max_sectors);
1162
1163 r1_bio->bios[rdisk] = read_bio;
1164
1165 read_bio->bi_iter.bi_sector = r1_bio->sector +
1166 mirror->rdev->data_offset;
1167 read_bio->bi_bdev = mirror->rdev->bdev;
1168 read_bio->bi_end_io = raid1_end_read_request;
1169 read_bio->bi_rw = READ | do_sync;
1170 read_bio->bi_private = r1_bio;
1171
1172 if (max_sectors < r1_bio->sectors) {
1173
1174
1175
1176
1177 sectors_handled = (r1_bio->sector + max_sectors
1178 - bio->bi_iter.bi_sector);
1179 r1_bio->sectors = max_sectors;
1180 spin_lock_irq(&conf->device_lock);
1181 if (bio->bi_phys_segments == 0)
1182 bio->bi_phys_segments = 2;
1183 else
1184 bio->bi_phys_segments++;
1185 spin_unlock_irq(&conf->device_lock);
1186
1187
1188
1189
1190
1191 reschedule_retry(r1_bio);
1192
1193 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1194
1195 r1_bio->master_bio = bio;
1196 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1197 r1_bio->state = 0;
1198 r1_bio->mddev = mddev;
1199 r1_bio->sector = bio->bi_iter.bi_sector +
1200 sectors_handled;
1201 goto read_again;
1202 } else
1203 generic_make_request(read_bio);
1204 return;
1205 }
1206
1207
1208
1209
1210 if (conf->pending_count >= max_queued_requests) {
1211 md_wakeup_thread(mddev->thread);
1212 wait_event(conf->wait_barrier,
1213 conf->pending_count < max_queued_requests);
1214 }
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226 disks = conf->raid_disks * 2;
1227 retry_write:
1228 r1_bio->start_next_window = start_next_window;
1229 blocked_rdev = NULL;
1230 rcu_read_lock();
1231 max_sectors = r1_bio->sectors;
1232 for (i = 0; i < disks; i++) {
1233 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1234 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1235 atomic_inc(&rdev->nr_pending);
1236 blocked_rdev = rdev;
1237 break;
1238 }
1239 r1_bio->bios[i] = NULL;
1240 if (!rdev || test_bit(Faulty, &rdev->flags)) {
1241 if (i < conf->raid_disks)
1242 set_bit(R1BIO_Degraded, &r1_bio->state);
1243 continue;
1244 }
1245
1246 atomic_inc(&rdev->nr_pending);
1247 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1248 sector_t first_bad;
1249 int bad_sectors;
1250 int is_bad;
1251
1252 is_bad = is_badblock(rdev, r1_bio->sector,
1253 max_sectors,
1254 &first_bad, &bad_sectors);
1255 if (is_bad < 0) {
1256
1257
1258 set_bit(BlockedBadBlocks, &rdev->flags);
1259 blocked_rdev = rdev;
1260 break;
1261 }
1262 if (is_bad && first_bad <= r1_bio->sector) {
1263
1264 bad_sectors -= (r1_bio->sector - first_bad);
1265 if (bad_sectors < max_sectors)
1266
1267
1268
1269 max_sectors = bad_sectors;
1270 rdev_dec_pending(rdev, mddev);
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281 continue;
1282 }
1283 if (is_bad) {
1284 int good_sectors = first_bad - r1_bio->sector;
1285 if (good_sectors < max_sectors)
1286 max_sectors = good_sectors;
1287 }
1288 }
1289 r1_bio->bios[i] = bio;
1290 }
1291 rcu_read_unlock();
1292
1293 if (unlikely(blocked_rdev)) {
1294
1295 int j;
1296 sector_t old = start_next_window;
1297
1298 for (j = 0; j < i; j++)
1299 if (r1_bio->bios[j])
1300 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
1301 r1_bio->state = 0;
1302 allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector);
1303 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1304 start_next_window = wait_barrier(conf, bio);
1305
1306
1307
1308
1309 if (bio->bi_phys_segments && old &&
1310 old != start_next_window)
1311
1312 wait_event(conf->wait_barrier,
1313 bio->bi_phys_segments == 1);
1314 goto retry_write;
1315 }
1316
1317 if (max_sectors < r1_bio->sectors) {
1318
1319
1320
1321 r1_bio->sectors = max_sectors;
1322 spin_lock_irq(&conf->device_lock);
1323 if (bio->bi_phys_segments == 0)
1324 bio->bi_phys_segments = 2;
1325 else
1326 bio->bi_phys_segments++;
1327 spin_unlock_irq(&conf->device_lock);
1328 }
1329 sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector;
1330
1331 atomic_set(&r1_bio->remaining, 1);
1332 atomic_set(&r1_bio->behind_remaining, 0);
1333
1334 first_clone = 1;
1335 for (i = 0; i < disks; i++) {
1336 struct bio *mbio;
1337 if (!r1_bio->bios[i])
1338 continue;
1339
1340 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1341 bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors);
1342
1343 if (first_clone) {
1344
1345
1346
1347
1348 if (bitmap &&
1349 (atomic_read(&bitmap->behind_writes)
1350 < mddev->bitmap_info.max_write_behind) &&
1351 !waitqueue_active(&bitmap->behind_wait))
1352 alloc_behind_pages(mbio, r1_bio);
1353
1354 bitmap_startwrite(bitmap, r1_bio->sector,
1355 r1_bio->sectors,
1356 test_bit(R1BIO_BehindIO,
1357 &r1_bio->state));
1358 first_clone = 0;
1359 }
1360 if (r1_bio->behind_bvecs) {
1361 struct bio_vec *bvec;
1362 int j;
1363
1364
1365
1366
1367 bio_for_each_segment_all(bvec, mbio, j)
1368 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
1369 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1370 atomic_inc(&r1_bio->behind_remaining);
1371 }
1372
1373 r1_bio->bios[i] = mbio;
1374
1375 mbio->bi_iter.bi_sector = (r1_bio->sector +
1376 conf->mirrors[i].rdev->data_offset);
1377 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1378 mbio->bi_end_io = raid1_end_write_request;
1379 mbio->bi_rw =
1380 WRITE | do_flush_fua | do_sync | do_discard | do_same;
1381 mbio->bi_private = r1_bio;
1382
1383 atomic_inc(&r1_bio->remaining);
1384
1385 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1386 if (cb)
1387 plug = container_of(cb, struct raid1_plug_cb, cb);
1388 else
1389 plug = NULL;
1390 spin_lock_irqsave(&conf->device_lock, flags);
1391 if (plug) {
1392 bio_list_add(&plug->pending, mbio);
1393 plug->pending_cnt++;
1394 } else {
1395 bio_list_add(&conf->pending_bio_list, mbio);
1396 conf->pending_count++;
1397 }
1398 spin_unlock_irqrestore(&conf->device_lock, flags);
1399 if (!plug)
1400 md_wakeup_thread(mddev->thread);
1401 }
1402
1403
1404
1405 if (sectors_handled < bio_sectors(bio)) {
1406 r1_bio_write_done(r1_bio);
1407
1408
1409
1410 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1411 r1_bio->master_bio = bio;
1412 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1413 r1_bio->state = 0;
1414 r1_bio->mddev = mddev;
1415 r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
1416 goto retry_write;
1417 }
1418
1419 r1_bio_write_done(r1_bio);
1420
1421
1422 wake_up(&conf->wait_barrier);
1423}
1424
1425static void raid1_status(struct seq_file *seq, struct mddev *mddev)
1426{
1427 struct r1conf *conf = mddev->private;
1428 int i;
1429
1430 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
1431 conf->raid_disks - mddev->degraded);
1432 rcu_read_lock();
1433 for (i = 0; i < conf->raid_disks; i++) {
1434 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1435 seq_printf(seq, "%s",
1436 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1437 }
1438 rcu_read_unlock();
1439 seq_printf(seq, "]");
1440}
1441
1442static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
1443{
1444 char b[BDEVNAME_SIZE];
1445 struct r1conf *conf = mddev->private;
1446 unsigned long flags;
1447
1448
1449
1450
1451
1452
1453
1454 if (test_bit(In_sync, &rdev->flags)
1455 && (conf->raid_disks - mddev->degraded) == 1) {
1456
1457
1458
1459
1460
1461
1462 conf->recovery_disabled = mddev->recovery_disabled;
1463 return;
1464 }
1465 set_bit(Blocked, &rdev->flags);
1466 spin_lock_irqsave(&conf->device_lock, flags);
1467 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1468 mddev->degraded++;
1469 set_bit(Faulty, &rdev->flags);
1470 } else
1471 set_bit(Faulty, &rdev->flags);
1472 spin_unlock_irqrestore(&conf->device_lock, flags);
1473
1474
1475
1476 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1477 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1478 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1479 printk(KERN_ALERT
1480 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1481 "md/raid1:%s: Operation continuing on %d devices.\n",
1482 mdname(mddev), bdevname(rdev->bdev, b),
1483 mdname(mddev), conf->raid_disks - mddev->degraded);
1484}
1485
1486static void print_conf(struct r1conf *conf)
1487{
1488 int i;
1489
1490 printk(KERN_DEBUG "RAID1 conf printout:\n");
1491 if (!conf) {
1492 printk(KERN_DEBUG "(!conf)\n");
1493 return;
1494 }
1495 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1496 conf->raid_disks);
1497
1498 rcu_read_lock();
1499 for (i = 0; i < conf->raid_disks; i++) {
1500 char b[BDEVNAME_SIZE];
1501 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1502 if (rdev)
1503 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
1504 i, !test_bit(In_sync, &rdev->flags),
1505 !test_bit(Faulty, &rdev->flags),
1506 bdevname(rdev->bdev,b));
1507 }
1508 rcu_read_unlock();
1509}
1510
1511static void close_sync(struct r1conf *conf)
1512{
1513 wait_barrier(conf, NULL);
1514 allow_barrier(conf, 0, 0);
1515
1516 mempool_destroy(conf->r1buf_pool);
1517 conf->r1buf_pool = NULL;
1518
1519 spin_lock_irq(&conf->resync_lock);
1520 conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE;
1521 conf->start_next_window = MaxSector;
1522 conf->current_window_requests +=
1523 conf->next_window_requests;
1524 conf->next_window_requests = 0;
1525 spin_unlock_irq(&conf->resync_lock);
1526}
1527
1528static int raid1_spare_active(struct mddev *mddev)
1529{
1530 int i;
1531 struct r1conf *conf = mddev->private;
1532 int count = 0;
1533 unsigned long flags;
1534
1535
1536
1537
1538
1539
1540
1541
1542 spin_lock_irqsave(&conf->device_lock, flags);
1543 for (i = 0; i < conf->raid_disks; i++) {
1544 struct md_rdev *rdev = conf->mirrors[i].rdev;
1545 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1546 if (repl
1547 && !test_bit(Candidate, &repl->flags)
1548 && repl->recovery_offset == MaxSector
1549 && !test_bit(Faulty, &repl->flags)
1550 && !test_and_set_bit(In_sync, &repl->flags)) {
1551
1552 if (!rdev ||
1553 !test_and_clear_bit(In_sync, &rdev->flags))
1554 count++;
1555 if (rdev) {
1556
1557
1558
1559
1560 set_bit(Faulty, &rdev->flags);
1561 sysfs_notify_dirent_safe(
1562 rdev->sysfs_state);
1563 }
1564 }
1565 if (rdev
1566 && rdev->recovery_offset == MaxSector
1567 && !test_bit(Faulty, &rdev->flags)
1568 && !test_and_set_bit(In_sync, &rdev->flags)) {
1569 count++;
1570 sysfs_notify_dirent_safe(rdev->sysfs_state);
1571 }
1572 }
1573 mddev->degraded -= count;
1574 spin_unlock_irqrestore(&conf->device_lock, flags);
1575
1576 print_conf(conf);
1577 return count;
1578}
1579
1580static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1581{
1582 struct r1conf *conf = mddev->private;
1583 int err = -EEXIST;
1584 int mirror = 0;
1585 struct raid1_info *p;
1586 int first = 0;
1587 int last = conf->raid_disks - 1;
1588
1589 if (mddev->recovery_disabled == conf->recovery_disabled)
1590 return -EBUSY;
1591
1592 if (md_integrity_add_rdev(rdev, mddev))
1593 return -ENXIO;
1594
1595 if (rdev->raid_disk >= 0)
1596 first = last = rdev->raid_disk;
1597
1598
1599
1600
1601
1602 if (rdev->saved_raid_disk >= 0 &&
1603 rdev->saved_raid_disk >= first &&
1604 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1605 first = last = rdev->saved_raid_disk;
1606
1607 for (mirror = first; mirror <= last; mirror++) {
1608 p = conf->mirrors+mirror;
1609 if (!p->rdev) {
1610
1611 if (mddev->gendisk)
1612 disk_stack_limits(mddev->gendisk, rdev->bdev,
1613 rdev->data_offset << 9);
1614
1615 p->head_position = 0;
1616 rdev->raid_disk = mirror;
1617 err = 0;
1618
1619
1620
1621 if (rdev->saved_raid_disk < 0)
1622 conf->fullsync = 1;
1623 rcu_assign_pointer(p->rdev, rdev);
1624 break;
1625 }
1626 if (test_bit(WantReplacement, &p->rdev->flags) &&
1627 p[conf->raid_disks].rdev == NULL) {
1628
1629 clear_bit(In_sync, &rdev->flags);
1630 set_bit(Replacement, &rdev->flags);
1631 rdev->raid_disk = mirror;
1632 err = 0;
1633 conf->fullsync = 1;
1634 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1635 break;
1636 }
1637 }
1638 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
1639 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
1640 print_conf(conf);
1641 return err;
1642}
1643
1644static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1645{
1646 struct r1conf *conf = mddev->private;
1647 int err = 0;
1648 int number = rdev->raid_disk;
1649 struct raid1_info *p = conf->mirrors + number;
1650
1651 if (rdev != p->rdev)
1652 p = conf->mirrors + conf->raid_disks + number;
1653
1654 print_conf(conf);
1655 if (rdev == p->rdev) {
1656 if (test_bit(In_sync, &rdev->flags) ||
1657 atomic_read(&rdev->nr_pending)) {
1658 err = -EBUSY;
1659 goto abort;
1660 }
1661
1662
1663
1664 if (!test_bit(Faulty, &rdev->flags) &&
1665 mddev->recovery_disabled != conf->recovery_disabled &&
1666 mddev->degraded < conf->raid_disks) {
1667 err = -EBUSY;
1668 goto abort;
1669 }
1670 p->rdev = NULL;
1671 synchronize_rcu();
1672 if (atomic_read(&rdev->nr_pending)) {
1673
1674 err = -EBUSY;
1675 p->rdev = rdev;
1676 goto abort;
1677 } else if (conf->mirrors[conf->raid_disks + number].rdev) {
1678
1679
1680
1681
1682 struct md_rdev *repl =
1683 conf->mirrors[conf->raid_disks + number].rdev;
1684 freeze_array(conf, 0);
1685 clear_bit(Replacement, &repl->flags);
1686 p->rdev = repl;
1687 conf->mirrors[conf->raid_disks + number].rdev = NULL;
1688 unfreeze_array(conf);
1689 clear_bit(WantReplacement, &rdev->flags);
1690 } else
1691 clear_bit(WantReplacement, &rdev->flags);
1692 err = md_integrity_register(mddev);
1693 }
1694abort:
1695
1696 print_conf(conf);
1697 return err;
1698}
1699
1700static void end_sync_read(struct bio *bio)
1701{
1702 struct r1bio *r1_bio = bio->bi_private;
1703
1704 update_head_pos(r1_bio->read_disk, r1_bio);
1705
1706
1707
1708
1709
1710
1711 if (!bio->bi_error)
1712 set_bit(R1BIO_Uptodate, &r1_bio->state);
1713
1714 if (atomic_dec_and_test(&r1_bio->remaining))
1715 reschedule_retry(r1_bio);
1716}
1717
1718static void end_sync_write(struct bio *bio)
1719{
1720 int uptodate = !bio->bi_error;
1721 struct r1bio *r1_bio = bio->bi_private;
1722 struct mddev *mddev = r1_bio->mddev;
1723 struct r1conf *conf = mddev->private;
1724 int mirror=0;
1725 sector_t first_bad;
1726 int bad_sectors;
1727
1728 mirror = find_bio_disk(r1_bio, bio);
1729
1730 if (!uptodate) {
1731 sector_t sync_blocks = 0;
1732 sector_t s = r1_bio->sector;
1733 long sectors_to_go = r1_bio->sectors;
1734
1735 do {
1736 bitmap_end_sync(mddev->bitmap, s,
1737 &sync_blocks, 1);
1738 s += sync_blocks;
1739 sectors_to_go -= sync_blocks;
1740 } while (sectors_to_go > 0);
1741 set_bit(WriteErrorSeen,
1742 &conf->mirrors[mirror].rdev->flags);
1743 if (!test_and_set_bit(WantReplacement,
1744 &conf->mirrors[mirror].rdev->flags))
1745 set_bit(MD_RECOVERY_NEEDED, &
1746 mddev->recovery);
1747 set_bit(R1BIO_WriteError, &r1_bio->state);
1748 } else if (is_badblock(conf->mirrors[mirror].rdev,
1749 r1_bio->sector,
1750 r1_bio->sectors,
1751 &first_bad, &bad_sectors) &&
1752 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1753 r1_bio->sector,
1754 r1_bio->sectors,
1755 &first_bad, &bad_sectors)
1756 )
1757 set_bit(R1BIO_MadeGood, &r1_bio->state);
1758
1759 if (atomic_dec_and_test(&r1_bio->remaining)) {
1760 int s = r1_bio->sectors;
1761 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1762 test_bit(R1BIO_WriteError, &r1_bio->state))
1763 reschedule_retry(r1_bio);
1764 else {
1765 put_buf(r1_bio);
1766 md_done_sync(mddev, s, uptodate);
1767 }
1768 }
1769}
1770
1771static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
1772 int sectors, struct page *page, int rw)
1773{
1774 if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
1775
1776 return 1;
1777 if (rw == WRITE) {
1778 set_bit(WriteErrorSeen, &rdev->flags);
1779 if (!test_and_set_bit(WantReplacement,
1780 &rdev->flags))
1781 set_bit(MD_RECOVERY_NEEDED, &
1782 rdev->mddev->recovery);
1783 }
1784
1785 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1786 md_error(rdev->mddev, rdev);
1787 return 0;
1788}
1789
1790static int fix_sync_read_error(struct r1bio *r1_bio)
1791{
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803 struct mddev *mddev = r1_bio->mddev;
1804 struct r1conf *conf = mddev->private;
1805 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1806 sector_t sect = r1_bio->sector;
1807 int sectors = r1_bio->sectors;
1808 int idx = 0;
1809
1810 while(sectors) {
1811 int s = sectors;
1812 int d = r1_bio->read_disk;
1813 int success = 0;
1814 struct md_rdev *rdev;
1815 int start;
1816
1817 if (s > (PAGE_SIZE>>9))
1818 s = PAGE_SIZE >> 9;
1819 do {
1820 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1821
1822
1823
1824
1825 rdev = conf->mirrors[d].rdev;
1826 if (sync_page_io(rdev, sect, s<<9,
1827 bio->bi_io_vec[idx].bv_page,
1828 READ, false)) {
1829 success = 1;
1830 break;
1831 }
1832 }
1833 d++;
1834 if (d == conf->raid_disks * 2)
1835 d = 0;
1836 } while (!success && d != r1_bio->read_disk);
1837
1838 if (!success) {
1839 char b[BDEVNAME_SIZE];
1840 int abort = 0;
1841
1842
1843
1844
1845
1846 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error"
1847 " for block %llu\n",
1848 mdname(mddev),
1849 bdevname(bio->bi_bdev, b),
1850 (unsigned long long)r1_bio->sector);
1851 for (d = 0; d < conf->raid_disks * 2; d++) {
1852 rdev = conf->mirrors[d].rdev;
1853 if (!rdev || test_bit(Faulty, &rdev->flags))
1854 continue;
1855 if (!rdev_set_badblocks(rdev, sect, s, 0))
1856 abort = 1;
1857 }
1858 if (abort) {
1859 conf->recovery_disabled =
1860 mddev->recovery_disabled;
1861 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1862 md_done_sync(mddev, r1_bio->sectors, 0);
1863 put_buf(r1_bio);
1864 return 0;
1865 }
1866
1867 sectors -= s;
1868 sect += s;
1869 idx++;
1870 continue;
1871 }
1872
1873 start = d;
1874
1875 while (d != r1_bio->read_disk) {
1876 if (d == 0)
1877 d = conf->raid_disks * 2;
1878 d--;
1879 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1880 continue;
1881 rdev = conf->mirrors[d].rdev;
1882 if (r1_sync_page_io(rdev, sect, s,
1883 bio->bi_io_vec[idx].bv_page,
1884 WRITE) == 0) {
1885 r1_bio->bios[d]->bi_end_io = NULL;
1886 rdev_dec_pending(rdev, mddev);
1887 }
1888 }
1889 d = start;
1890 while (d != r1_bio->read_disk) {
1891 if (d == 0)
1892 d = conf->raid_disks * 2;
1893 d--;
1894 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1895 continue;
1896 rdev = conf->mirrors[d].rdev;
1897 if (r1_sync_page_io(rdev, sect, s,
1898 bio->bi_io_vec[idx].bv_page,
1899 READ) != 0)
1900 atomic_add(s, &rdev->corrected_errors);
1901 }
1902 sectors -= s;
1903 sect += s;
1904 idx ++;
1905 }
1906 set_bit(R1BIO_Uptodate, &r1_bio->state);
1907 bio->bi_error = 0;
1908 return 1;
1909}
1910
1911static void process_checks(struct r1bio *r1_bio)
1912{
1913
1914
1915
1916
1917
1918
1919
1920 struct mddev *mddev = r1_bio->mddev;
1921 struct r1conf *conf = mddev->private;
1922 int primary;
1923 int i;
1924 int vcnt;
1925
1926
1927 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
1928 for (i = 0; i < conf->raid_disks * 2; i++) {
1929 int j;
1930 int size;
1931 int error;
1932 struct bio *b = r1_bio->bios[i];
1933 if (b->bi_end_io != end_sync_read)
1934 continue;
1935
1936 error = b->bi_error;
1937 bio_reset(b);
1938 b->bi_error = error;
1939 b->bi_vcnt = vcnt;
1940 b->bi_iter.bi_size = r1_bio->sectors << 9;
1941 b->bi_iter.bi_sector = r1_bio->sector +
1942 conf->mirrors[i].rdev->data_offset;
1943 b->bi_bdev = conf->mirrors[i].rdev->bdev;
1944 b->bi_end_io = end_sync_read;
1945 b->bi_private = r1_bio;
1946
1947 size = b->bi_iter.bi_size;
1948 for (j = 0; j < vcnt ; j++) {
1949 struct bio_vec *bi;
1950 bi = &b->bi_io_vec[j];
1951 bi->bv_offset = 0;
1952 if (size > PAGE_SIZE)
1953 bi->bv_len = PAGE_SIZE;
1954 else
1955 bi->bv_len = size;
1956 size -= PAGE_SIZE;
1957 }
1958 }
1959 for (primary = 0; primary < conf->raid_disks * 2; primary++)
1960 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1961 !r1_bio->bios[primary]->bi_error) {
1962 r1_bio->bios[primary]->bi_end_io = NULL;
1963 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1964 break;
1965 }
1966 r1_bio->read_disk = primary;
1967 for (i = 0; i < conf->raid_disks * 2; i++) {
1968 int j;
1969 struct bio *pbio = r1_bio->bios[primary];
1970 struct bio *sbio = r1_bio->bios[i];
1971 int error = sbio->bi_error;
1972
1973 if (sbio->bi_end_io != end_sync_read)
1974 continue;
1975
1976 sbio->bi_error = 0;
1977
1978 if (!error) {
1979 for (j = vcnt; j-- ; ) {
1980 struct page *p, *s;
1981 p = pbio->bi_io_vec[j].bv_page;
1982 s = sbio->bi_io_vec[j].bv_page;
1983 if (memcmp(page_address(p),
1984 page_address(s),
1985 sbio->bi_io_vec[j].bv_len))
1986 break;
1987 }
1988 } else
1989 j = 0;
1990 if (j >= 0)
1991 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
1992 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
1993 && !error)) {
1994
1995 sbio->bi_end_io = NULL;
1996 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1997 continue;
1998 }
1999
2000 bio_copy_data(sbio, pbio);
2001 }
2002}
2003
2004static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
2005{
2006 struct r1conf *conf = mddev->private;
2007 int i;
2008 int disks = conf->raid_disks * 2;
2009 struct bio *bio, *wbio;
2010
2011 bio = r1_bio->bios[r1_bio->read_disk];
2012
2013 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
2014
2015 if (!fix_sync_read_error(r1_bio))
2016 return;
2017
2018 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2019 process_checks(r1_bio);
2020
2021
2022
2023
2024 atomic_set(&r1_bio->remaining, 1);
2025 for (i = 0; i < disks ; i++) {
2026 wbio = r1_bio->bios[i];
2027 if (wbio->bi_end_io == NULL ||
2028 (wbio->bi_end_io == end_sync_read &&
2029 (i == r1_bio->read_disk ||
2030 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
2031 continue;
2032
2033 wbio->bi_rw = WRITE;
2034 wbio->bi_end_io = end_sync_write;
2035 atomic_inc(&r1_bio->remaining);
2036 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
2037
2038 generic_make_request(wbio);
2039 }
2040
2041 if (atomic_dec_and_test(&r1_bio->remaining)) {
2042
2043 int s = r1_bio->sectors;
2044 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2045 test_bit(R1BIO_WriteError, &r1_bio->state))
2046 reschedule_retry(r1_bio);
2047 else {
2048 put_buf(r1_bio);
2049 md_done_sync(mddev, s, 1);
2050 }
2051 }
2052}
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062static void fix_read_error(struct r1conf *conf, int read_disk,
2063 sector_t sect, int sectors)
2064{
2065 struct mddev *mddev = conf->mddev;
2066 while(sectors) {
2067 int s = sectors;
2068 int d = read_disk;
2069 int success = 0;
2070 int start;
2071 struct md_rdev *rdev;
2072
2073 if (s > (PAGE_SIZE>>9))
2074 s = PAGE_SIZE >> 9;
2075
2076 do {
2077
2078
2079
2080
2081
2082 sector_t first_bad;
2083 int bad_sectors;
2084
2085 rdev = conf->mirrors[d].rdev;
2086 if (rdev &&
2087 (test_bit(In_sync, &rdev->flags) ||
2088 (!test_bit(Faulty, &rdev->flags) &&
2089 rdev->recovery_offset >= sect + s)) &&
2090 is_badblock(rdev, sect, s,
2091 &first_bad, &bad_sectors) == 0 &&
2092 sync_page_io(rdev, sect, s<<9,
2093 conf->tmppage, READ, false))
2094 success = 1;
2095 else {
2096 d++;
2097 if (d == conf->raid_disks * 2)
2098 d = 0;
2099 }
2100 } while (!success && d != read_disk);
2101
2102 if (!success) {
2103
2104 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
2105 if (!rdev_set_badblocks(rdev, sect, s, 0))
2106 md_error(mddev, rdev);
2107 break;
2108 }
2109
2110 start = d;
2111 while (d != read_disk) {
2112 if (d==0)
2113 d = conf->raid_disks * 2;
2114 d--;
2115 rdev = conf->mirrors[d].rdev;
2116 if (rdev &&
2117 !test_bit(Faulty, &rdev->flags))
2118 r1_sync_page_io(rdev, sect, s,
2119 conf->tmppage, WRITE);
2120 }
2121 d = start;
2122 while (d != read_disk) {
2123 char b[BDEVNAME_SIZE];
2124 if (d==0)
2125 d = conf->raid_disks * 2;
2126 d--;
2127 rdev = conf->mirrors[d].rdev;
2128 if (rdev &&
2129 !test_bit(Faulty, &rdev->flags)) {
2130 if (r1_sync_page_io(rdev, sect, s,
2131 conf->tmppage, READ)) {
2132 atomic_add(s, &rdev->corrected_errors);
2133 printk(KERN_INFO
2134 "md/raid1:%s: read error corrected "
2135 "(%d sectors at %llu on %s)\n",
2136 mdname(mddev), s,
2137 (unsigned long long)(sect +
2138 rdev->data_offset),
2139 bdevname(rdev->bdev, b));
2140 }
2141 }
2142 }
2143 sectors -= s;
2144 sect += s;
2145 }
2146}
2147
2148static int narrow_write_error(struct r1bio *r1_bio, int i)
2149{
2150 struct mddev *mddev = r1_bio->mddev;
2151 struct r1conf *conf = mddev->private;
2152 struct md_rdev *rdev = conf->mirrors[i].rdev;
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165 int block_sectors;
2166 sector_t sector;
2167 int sectors;
2168 int sect_to_write = r1_bio->sectors;
2169 int ok = 1;
2170
2171 if (rdev->badblocks.shift < 0)
2172 return 0;
2173
2174 block_sectors = roundup(1 << rdev->badblocks.shift,
2175 bdev_logical_block_size(rdev->bdev) >> 9);
2176 sector = r1_bio->sector;
2177 sectors = ((sector + block_sectors)
2178 & ~(sector_t)(block_sectors - 1))
2179 - sector;
2180
2181 while (sect_to_write) {
2182 struct bio *wbio;
2183 if (sectors > sect_to_write)
2184 sectors = sect_to_write;
2185
2186
2187 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2188 unsigned vcnt = r1_bio->behind_page_count;
2189 struct bio_vec *vec = r1_bio->behind_bvecs;
2190
2191 while (!vec->bv_page) {
2192 vec++;
2193 vcnt--;
2194 }
2195
2196 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
2197 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
2198
2199 wbio->bi_vcnt = vcnt;
2200 } else {
2201 wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2202 }
2203
2204 wbio->bi_rw = WRITE;
2205 wbio->bi_iter.bi_sector = r1_bio->sector;
2206 wbio->bi_iter.bi_size = r1_bio->sectors << 9;
2207
2208 bio_trim(wbio, sector - r1_bio->sector, sectors);
2209 wbio->bi_iter.bi_sector += rdev->data_offset;
2210 wbio->bi_bdev = rdev->bdev;
2211 if (submit_bio_wait(WRITE, wbio) < 0)
2212
2213 ok = rdev_set_badblocks(rdev, sector,
2214 sectors, 0)
2215 && ok;
2216
2217 bio_put(wbio);
2218 sect_to_write -= sectors;
2219 sector += sectors;
2220 sectors = block_sectors;
2221 }
2222 return ok;
2223}
2224
2225static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2226{
2227 int m;
2228 int s = r1_bio->sectors;
2229 for (m = 0; m < conf->raid_disks * 2 ; m++) {
2230 struct md_rdev *rdev = conf->mirrors[m].rdev;
2231 struct bio *bio = r1_bio->bios[m];
2232 if (bio->bi_end_io == NULL)
2233 continue;
2234 if (!bio->bi_error &&
2235 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
2236 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
2237 }
2238 if (bio->bi_error &&
2239 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2240 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2241 md_error(conf->mddev, rdev);
2242 }
2243 }
2244 put_buf(r1_bio);
2245 md_done_sync(conf->mddev, s, 1);
2246}
2247
2248static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2249{
2250 int m;
2251 bool fail = false;
2252 for (m = 0; m < conf->raid_disks * 2 ; m++)
2253 if (r1_bio->bios[m] == IO_MADE_GOOD) {
2254 struct md_rdev *rdev = conf->mirrors[m].rdev;
2255 rdev_clear_badblocks(rdev,
2256 r1_bio->sector,
2257 r1_bio->sectors, 0);
2258 rdev_dec_pending(rdev, conf->mddev);
2259 } else if (r1_bio->bios[m] != NULL) {
2260
2261
2262
2263
2264 fail = true;
2265 if (!narrow_write_error(r1_bio, m)) {
2266 md_error(conf->mddev,
2267 conf->mirrors[m].rdev);
2268
2269 set_bit(R1BIO_Degraded, &r1_bio->state);
2270 }
2271 rdev_dec_pending(conf->mirrors[m].rdev,
2272 conf->mddev);
2273 }
2274 if (fail) {
2275 spin_lock_irq(&conf->device_lock);
2276 list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
2277 spin_unlock_irq(&conf->device_lock);
2278 md_wakeup_thread(conf->mddev->thread);
2279 } else {
2280 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2281 close_write(r1_bio);
2282 raid_end_bio_io(r1_bio);
2283 }
2284}
2285
2286static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
2287{
2288 int disk;
2289 int max_sectors;
2290 struct mddev *mddev = conf->mddev;
2291 struct bio *bio;
2292 char b[BDEVNAME_SIZE];
2293 struct md_rdev *rdev;
2294
2295 clear_bit(R1BIO_ReadError, &r1_bio->state);
2296
2297
2298
2299
2300
2301
2302
2303
2304 if (mddev->ro == 0) {
2305 freeze_array(conf, 1);
2306 fix_read_error(conf, r1_bio->read_disk,
2307 r1_bio->sector, r1_bio->sectors);
2308 unfreeze_array(conf);
2309 } else
2310 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
2311 rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev);
2312
2313 bio = r1_bio->bios[r1_bio->read_disk];
2314 bdevname(bio->bi_bdev, b);
2315read_more:
2316 disk = read_balance(conf, r1_bio, &max_sectors);
2317 if (disk == -1) {
2318 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O"
2319 " read error for block %llu\n",
2320 mdname(mddev), b, (unsigned long long)r1_bio->sector);
2321 raid_end_bio_io(r1_bio);
2322 } else {
2323 const unsigned long do_sync
2324 = r1_bio->master_bio->bi_rw & REQ_SYNC;
2325 if (bio) {
2326 r1_bio->bios[r1_bio->read_disk] =
2327 mddev->ro ? IO_BLOCKED : NULL;
2328 bio_put(bio);
2329 }
2330 r1_bio->read_disk = disk;
2331 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2332 bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector,
2333 max_sectors);
2334 r1_bio->bios[r1_bio->read_disk] = bio;
2335 rdev = conf->mirrors[disk].rdev;
2336 printk_ratelimited(KERN_ERR
2337 "md/raid1:%s: redirecting sector %llu"
2338 " to other mirror: %s\n",
2339 mdname(mddev),
2340 (unsigned long long)r1_bio->sector,
2341 bdevname(rdev->bdev, b));
2342 bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset;
2343 bio->bi_bdev = rdev->bdev;
2344 bio->bi_end_io = raid1_end_read_request;
2345 bio->bi_rw = READ | do_sync;
2346 bio->bi_private = r1_bio;
2347 if (max_sectors < r1_bio->sectors) {
2348
2349 struct bio *mbio = r1_bio->master_bio;
2350 int sectors_handled = (r1_bio->sector + max_sectors
2351 - mbio->bi_iter.bi_sector);
2352 r1_bio->sectors = max_sectors;
2353 spin_lock_irq(&conf->device_lock);
2354 if (mbio->bi_phys_segments == 0)
2355 mbio->bi_phys_segments = 2;
2356 else
2357 mbio->bi_phys_segments++;
2358 spin_unlock_irq(&conf->device_lock);
2359 generic_make_request(bio);
2360 bio = NULL;
2361
2362 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2363
2364 r1_bio->master_bio = mbio;
2365 r1_bio->sectors = bio_sectors(mbio) - sectors_handled;
2366 r1_bio->state = 0;
2367 set_bit(R1BIO_ReadError, &r1_bio->state);
2368 r1_bio->mddev = mddev;
2369 r1_bio->sector = mbio->bi_iter.bi_sector +
2370 sectors_handled;
2371
2372 goto read_more;
2373 } else
2374 generic_make_request(bio);
2375 }
2376}
2377
2378static void raid1d(struct md_thread *thread)
2379{
2380 struct mddev *mddev = thread->mddev;
2381 struct r1bio *r1_bio;
2382 unsigned long flags;
2383 struct r1conf *conf = mddev->private;
2384 struct list_head *head = &conf->retry_list;
2385 struct blk_plug plug;
2386
2387 md_check_recovery(mddev);
2388
2389 if (!list_empty_careful(&conf->bio_end_io_list) &&
2390 !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2391 LIST_HEAD(tmp);
2392 spin_lock_irqsave(&conf->device_lock, flags);
2393 if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2394 list_add(&tmp, &conf->bio_end_io_list);
2395 list_del_init(&conf->bio_end_io_list);
2396 }
2397 spin_unlock_irqrestore(&conf->device_lock, flags);
2398 while (!list_empty(&tmp)) {
2399 r1_bio = list_first_entry(&tmp, struct r1bio,
2400 retry_list);
2401 list_del(&r1_bio->retry_list);
2402 if (mddev->degraded)
2403 set_bit(R1BIO_Degraded, &r1_bio->state);
2404 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2405 close_write(r1_bio);
2406 raid_end_bio_io(r1_bio);
2407 }
2408 }
2409
2410 blk_start_plug(&plug);
2411 for (;;) {
2412
2413 flush_pending_writes(conf);
2414
2415 spin_lock_irqsave(&conf->device_lock, flags);
2416 if (list_empty(head)) {
2417 spin_unlock_irqrestore(&conf->device_lock, flags);
2418 break;
2419 }
2420 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
2421 list_del(head->prev);
2422 conf->nr_queued--;
2423 spin_unlock_irqrestore(&conf->device_lock, flags);
2424
2425 mddev = r1_bio->mddev;
2426 conf = mddev->private;
2427 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
2428 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2429 test_bit(R1BIO_WriteError, &r1_bio->state))
2430 handle_sync_write_finished(conf, r1_bio);
2431 else
2432 sync_request_write(mddev, r1_bio);
2433 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2434 test_bit(R1BIO_WriteError, &r1_bio->state))
2435 handle_write_finished(conf, r1_bio);
2436 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2437 handle_read_error(conf, r1_bio);
2438 else
2439
2440
2441
2442 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
2443
2444 cond_resched();
2445 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2446 md_check_recovery(mddev);
2447 }
2448 blk_finish_plug(&plug);
2449}
2450
2451static int init_resync(struct r1conf *conf)
2452{
2453 int buffs;
2454
2455 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
2456 BUG_ON(conf->r1buf_pool);
2457 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2458 conf->poolinfo);
2459 if (!conf->r1buf_pool)
2460 return -ENOMEM;
2461 conf->next_resync = 0;
2462 return 0;
2463}
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
2476 int *skipped)
2477{
2478 struct r1conf *conf = mddev->private;
2479 struct r1bio *r1_bio;
2480 struct bio *bio;
2481 sector_t max_sector, nr_sectors;
2482 int disk = -1;
2483 int i;
2484 int wonly = -1;
2485 int write_targets = 0, read_targets = 0;
2486 sector_t sync_blocks;
2487 int still_degraded = 0;
2488 int good_sectors = RESYNC_SECTORS;
2489 int min_bad = 0;
2490
2491 if (!conf->r1buf_pool)
2492 if (init_resync(conf))
2493 return 0;
2494
2495 max_sector = mddev->dev_sectors;
2496 if (sector_nr >= max_sector) {
2497
2498
2499
2500
2501
2502 if (mddev->curr_resync < max_sector)
2503 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
2504 &sync_blocks, 1);
2505 else
2506 conf->fullsync = 0;
2507
2508 bitmap_close_sync(mddev->bitmap);
2509 close_sync(conf);
2510
2511 if (mddev_is_clustered(mddev)) {
2512 conf->cluster_sync_low = 0;
2513 conf->cluster_sync_high = 0;
2514 }
2515 return 0;
2516 }
2517
2518 if (mddev->bitmap == NULL &&
2519 mddev->recovery_cp == MaxSector &&
2520 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
2521 conf->fullsync == 0) {
2522 *skipped = 1;
2523 return max_sector - sector_nr;
2524 }
2525
2526
2527
2528 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
2529 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2530
2531 *skipped = 1;
2532 return sync_blocks;
2533 }
2534
2535
2536
2537
2538
2539 bitmap_cond_end_sync(mddev->bitmap, sector_nr,
2540 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
2541 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
2542
2543 raise_barrier(conf, sector_nr);
2544
2545 rcu_read_lock();
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555 r1_bio->mddev = mddev;
2556 r1_bio->sector = sector_nr;
2557 r1_bio->state = 0;
2558 set_bit(R1BIO_IsSync, &r1_bio->state);
2559
2560 for (i = 0; i < conf->raid_disks * 2; i++) {
2561 struct md_rdev *rdev;
2562 bio = r1_bio->bios[i];
2563 bio_reset(bio);
2564
2565 rdev = rcu_dereference(conf->mirrors[i].rdev);
2566 if (rdev == NULL ||
2567 test_bit(Faulty, &rdev->flags)) {
2568 if (i < conf->raid_disks)
2569 still_degraded = 1;
2570 } else if (!test_bit(In_sync, &rdev->flags)) {
2571 bio->bi_rw = WRITE;
2572 bio->bi_end_io = end_sync_write;
2573 write_targets ++;
2574 } else {
2575
2576 sector_t first_bad = MaxSector;
2577 int bad_sectors;
2578
2579 if (is_badblock(rdev, sector_nr, good_sectors,
2580 &first_bad, &bad_sectors)) {
2581 if (first_bad > sector_nr)
2582 good_sectors = first_bad - sector_nr;
2583 else {
2584 bad_sectors -= (sector_nr - first_bad);
2585 if (min_bad == 0 ||
2586 min_bad > bad_sectors)
2587 min_bad = bad_sectors;
2588 }
2589 }
2590 if (sector_nr < first_bad) {
2591 if (test_bit(WriteMostly, &rdev->flags)) {
2592 if (wonly < 0)
2593 wonly = i;
2594 } else {
2595 if (disk < 0)
2596 disk = i;
2597 }
2598 bio->bi_rw = READ;
2599 bio->bi_end_io = end_sync_read;
2600 read_targets++;
2601 } else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2602 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2603 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2604
2605
2606
2607
2608
2609
2610 bio->bi_rw = WRITE;
2611 bio->bi_end_io = end_sync_write;
2612 write_targets++;
2613 }
2614 }
2615 if (bio->bi_end_io) {
2616 atomic_inc(&rdev->nr_pending);
2617 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
2618 bio->bi_bdev = rdev->bdev;
2619 bio->bi_private = r1_bio;
2620 }
2621 }
2622 rcu_read_unlock();
2623 if (disk < 0)
2624 disk = wonly;
2625 r1_bio->read_disk = disk;
2626
2627 if (read_targets == 0 && min_bad > 0) {
2628
2629
2630
2631 int ok = 1;
2632 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
2633 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
2634 struct md_rdev *rdev = conf->mirrors[i].rdev;
2635 ok = rdev_set_badblocks(rdev, sector_nr,
2636 min_bad, 0
2637 ) && ok;
2638 }
2639 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2640 *skipped = 1;
2641 put_buf(r1_bio);
2642
2643 if (!ok) {
2644
2645
2646
2647
2648
2649 conf->recovery_disabled = mddev->recovery_disabled;
2650 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2651 return 0;
2652 } else
2653 return min_bad;
2654
2655 }
2656 if (min_bad > 0 && min_bad < good_sectors) {
2657
2658
2659 good_sectors = min_bad;
2660 }
2661
2662 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2663
2664 write_targets += read_targets-1;
2665
2666 if (write_targets == 0 || read_targets == 0) {
2667
2668
2669
2670 sector_t rv;
2671 if (min_bad > 0)
2672 max_sector = sector_nr + min_bad;
2673 rv = max_sector - sector_nr;
2674 *skipped = 1;
2675 put_buf(r1_bio);
2676 return rv;
2677 }
2678
2679 if (max_sector > mddev->resync_max)
2680 max_sector = mddev->resync_max;
2681 if (max_sector > sector_nr + good_sectors)
2682 max_sector = sector_nr + good_sectors;
2683 nr_sectors = 0;
2684 sync_blocks = 0;
2685 do {
2686 struct page *page;
2687 int len = PAGE_SIZE;
2688 if (sector_nr + (len>>9) > max_sector)
2689 len = (max_sector - sector_nr) << 9;
2690 if (len == 0)
2691 break;
2692 if (sync_blocks == 0) {
2693 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
2694 &sync_blocks, still_degraded) &&
2695 !conf->fullsync &&
2696 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2697 break;
2698 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
2699 if ((len >> 9) > sync_blocks)
2700 len = sync_blocks<<9;
2701 }
2702
2703 for (i = 0 ; i < conf->raid_disks * 2; i++) {
2704 bio = r1_bio->bios[i];
2705 if (bio->bi_end_io) {
2706 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
2707 if (bio_add_page(bio, page, len, 0) == 0) {
2708
2709 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
2710 while (i > 0) {
2711 i--;
2712 bio = r1_bio->bios[i];
2713 if (bio->bi_end_io==NULL)
2714 continue;
2715
2716 bio->bi_vcnt--;
2717 bio->bi_iter.bi_size -= len;
2718 bio_clear_flag(bio, BIO_SEG_VALID);
2719 }
2720 goto bio_full;
2721 }
2722 }
2723 }
2724 nr_sectors += len>>9;
2725 sector_nr += len>>9;
2726 sync_blocks -= (len>>9);
2727 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2728 bio_full:
2729 r1_bio->sectors = nr_sectors;
2730
2731 if (mddev_is_clustered(mddev) &&
2732 conf->cluster_sync_high < sector_nr + nr_sectors) {
2733 conf->cluster_sync_low = mddev->curr_resync_completed;
2734 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
2735
2736 md_cluster_ops->resync_info_update(mddev,
2737 conf->cluster_sync_low,
2738 conf->cluster_sync_high);
2739 }
2740
2741
2742
2743
2744 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2745 atomic_set(&r1_bio->remaining, read_targets);
2746 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
2747 bio = r1_bio->bios[i];
2748 if (bio->bi_end_io == end_sync_read) {
2749 read_targets--;
2750 md_sync_acct(bio->bi_bdev, nr_sectors);
2751 generic_make_request(bio);
2752 }
2753 }
2754 } else {
2755 atomic_set(&r1_bio->remaining, 1);
2756 bio = r1_bio->bios[r1_bio->read_disk];
2757 md_sync_acct(bio->bi_bdev, nr_sectors);
2758 generic_make_request(bio);
2759
2760 }
2761 return nr_sectors;
2762}
2763
2764static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
2765{
2766 if (sectors)
2767 return sectors;
2768
2769 return mddev->dev_sectors;
2770}
2771
2772static struct r1conf *setup_conf(struct mddev *mddev)
2773{
2774 struct r1conf *conf;
2775 int i;
2776 struct raid1_info *disk;
2777 struct md_rdev *rdev;
2778 int err = -ENOMEM;
2779
2780 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
2781 if (!conf)
2782 goto abort;
2783
2784 conf->mirrors = kzalloc(sizeof(struct raid1_info)
2785 * mddev->raid_disks * 2,
2786 GFP_KERNEL);
2787 if (!conf->mirrors)
2788 goto abort;
2789
2790 conf->tmppage = alloc_page(GFP_KERNEL);
2791 if (!conf->tmppage)
2792 goto abort;
2793
2794 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
2795 if (!conf->poolinfo)
2796 goto abort;
2797 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
2798 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2799 r1bio_pool_free,
2800 conf->poolinfo);
2801 if (!conf->r1bio_pool)
2802 goto abort;
2803
2804 conf->poolinfo->mddev = mddev;
2805
2806 err = -EINVAL;
2807 spin_lock_init(&conf->device_lock);
2808 rdev_for_each(rdev, mddev) {
2809 struct request_queue *q;
2810 int disk_idx = rdev->raid_disk;
2811 if (disk_idx >= mddev->raid_disks
2812 || disk_idx < 0)
2813 continue;
2814 if (test_bit(Replacement, &rdev->flags))
2815 disk = conf->mirrors + mddev->raid_disks + disk_idx;
2816 else
2817 disk = conf->mirrors + disk_idx;
2818
2819 if (disk->rdev)
2820 goto abort;
2821 disk->rdev = rdev;
2822 q = bdev_get_queue(rdev->bdev);
2823
2824 disk->head_position = 0;
2825 disk->seq_start = MaxSector;
2826 }
2827 conf->raid_disks = mddev->raid_disks;
2828 conf->mddev = mddev;
2829 INIT_LIST_HEAD(&conf->retry_list);
2830 INIT_LIST_HEAD(&conf->bio_end_io_list);
2831
2832 spin_lock_init(&conf->resync_lock);
2833 init_waitqueue_head(&conf->wait_barrier);
2834
2835 bio_list_init(&conf->pending_bio_list);
2836 conf->pending_count = 0;
2837 conf->recovery_disabled = mddev->recovery_disabled - 1;
2838
2839 conf->start_next_window = MaxSector;
2840 conf->current_window_requests = conf->next_window_requests = 0;
2841
2842 err = -EIO;
2843 for (i = 0; i < conf->raid_disks * 2; i++) {
2844
2845 disk = conf->mirrors + i;
2846
2847 if (i < conf->raid_disks &&
2848 disk[conf->raid_disks].rdev) {
2849
2850 if (!disk->rdev) {
2851
2852
2853
2854 disk->rdev =
2855 disk[conf->raid_disks].rdev;
2856 disk[conf->raid_disks].rdev = NULL;
2857 } else if (!test_bit(In_sync, &disk->rdev->flags))
2858
2859 goto abort;
2860 }
2861
2862 if (!disk->rdev ||
2863 !test_bit(In_sync, &disk->rdev->flags)) {
2864 disk->head_position = 0;
2865 if (disk->rdev &&
2866 (disk->rdev->saved_raid_disk < 0))
2867 conf->fullsync = 1;
2868 }
2869 }
2870
2871 err = -ENOMEM;
2872 conf->thread = md_register_thread(raid1d, mddev, "raid1");
2873 if (!conf->thread) {
2874 printk(KERN_ERR
2875 "md/raid1:%s: couldn't allocate thread\n",
2876 mdname(mddev));
2877 goto abort;
2878 }
2879
2880 return conf;
2881
2882 abort:
2883 if (conf) {
2884 mempool_destroy(conf->r1bio_pool);
2885 kfree(conf->mirrors);
2886 safe_put_page(conf->tmppage);
2887 kfree(conf->poolinfo);
2888 kfree(conf);
2889 }
2890 return ERR_PTR(err);
2891}
2892
2893static void raid1_free(struct mddev *mddev, void *priv);
2894static int raid1_run(struct mddev *mddev)
2895{
2896 struct r1conf *conf;
2897 int i;
2898 struct md_rdev *rdev;
2899 int ret;
2900 bool discard_supported = false;
2901
2902 if (mddev->level != 1) {
2903 printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n",
2904 mdname(mddev), mddev->level);
2905 return -EIO;
2906 }
2907 if (mddev->reshape_position != MaxSector) {
2908 printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n",
2909 mdname(mddev));
2910 return -EIO;
2911 }
2912
2913
2914
2915
2916
2917 if (mddev->private == NULL)
2918 conf = setup_conf(mddev);
2919 else
2920 conf = mddev->private;
2921
2922 if (IS_ERR(conf))
2923 return PTR_ERR(conf);
2924
2925 if (mddev->queue)
2926 blk_queue_max_write_same_sectors(mddev->queue, 0);
2927
2928 rdev_for_each(rdev, mddev) {
2929 if (!mddev->gendisk)
2930 continue;
2931 disk_stack_limits(mddev->gendisk, rdev->bdev,
2932 rdev->data_offset << 9);
2933 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
2934 discard_supported = true;
2935 }
2936
2937 mddev->degraded = 0;
2938 for (i=0; i < conf->raid_disks; i++)
2939 if (conf->mirrors[i].rdev == NULL ||
2940 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2941 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2942 mddev->degraded++;
2943
2944 if (conf->raid_disks - mddev->degraded == 1)
2945 mddev->recovery_cp = MaxSector;
2946
2947 if (mddev->recovery_cp != MaxSector)
2948 printk(KERN_NOTICE "md/raid1:%s: not clean"
2949 " -- starting background reconstruction\n",
2950 mdname(mddev));
2951 printk(KERN_INFO
2952 "md/raid1:%s: active with %d out of %d mirrors\n",
2953 mdname(mddev), mddev->raid_disks - mddev->degraded,
2954 mddev->raid_disks);
2955
2956
2957
2958
2959 mddev->thread = conf->thread;
2960 conf->thread = NULL;
2961 mddev->private = conf;
2962
2963 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
2964
2965 if (mddev->queue) {
2966 if (discard_supported)
2967 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
2968 mddev->queue);
2969 else
2970 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
2971 mddev->queue);
2972 }
2973
2974 ret = md_integrity_register(mddev);
2975 if (ret) {
2976 md_unregister_thread(&mddev->thread);
2977 raid1_free(mddev, conf);
2978 }
2979 return ret;
2980}
2981
2982static void raid1_free(struct mddev *mddev, void *priv)
2983{
2984 struct r1conf *conf = priv;
2985
2986 mempool_destroy(conf->r1bio_pool);
2987 kfree(conf->mirrors);
2988 safe_put_page(conf->tmppage);
2989 kfree(conf->poolinfo);
2990 kfree(conf);
2991}
2992
2993static int raid1_resize(struct mddev *mddev, sector_t sectors)
2994{
2995
2996
2997
2998
2999
3000
3001
3002 sector_t newsize = raid1_size(mddev, sectors, 0);
3003 if (mddev->external_size &&
3004 mddev->array_sectors > newsize)
3005 return -EINVAL;
3006 if (mddev->bitmap) {
3007 int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
3008 if (ret)
3009 return ret;
3010 }
3011 md_set_array_sectors(mddev, newsize);
3012 set_capacity(mddev->gendisk, mddev->array_sectors);
3013 revalidate_disk(mddev->gendisk);
3014 if (sectors > mddev->dev_sectors &&
3015 mddev->recovery_cp > mddev->dev_sectors) {
3016 mddev->recovery_cp = mddev->dev_sectors;
3017 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3018 }
3019 mddev->dev_sectors = sectors;
3020 mddev->resync_max_sectors = sectors;
3021 return 0;
3022}
3023
3024static int raid1_reshape(struct mddev *mddev)
3025{
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037 mempool_t *newpool, *oldpool;
3038 struct pool_info *newpoolinfo;
3039 struct raid1_info *newmirrors;
3040 struct r1conf *conf = mddev->private;
3041 int cnt, raid_disks;
3042 unsigned long flags;
3043 int d, d2, err;
3044
3045
3046 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
3047 mddev->layout != mddev->new_layout ||
3048 mddev->level != mddev->new_level) {
3049 mddev->new_chunk_sectors = mddev->chunk_sectors;
3050 mddev->new_layout = mddev->layout;
3051 mddev->new_level = mddev->level;
3052 return -EINVAL;
3053 }
3054
3055 if (!mddev_is_clustered(mddev)) {
3056 err = md_allow_write(mddev);
3057 if (err)
3058 return err;
3059 }
3060
3061 raid_disks = mddev->raid_disks + mddev->delta_disks;
3062
3063 if (raid_disks < conf->raid_disks) {
3064 cnt=0;
3065 for (d= 0; d < conf->raid_disks; d++)
3066 if (conf->mirrors[d].rdev)
3067 cnt++;
3068 if (cnt > raid_disks)
3069 return -EBUSY;
3070 }
3071
3072 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
3073 if (!newpoolinfo)
3074 return -ENOMEM;
3075 newpoolinfo->mddev = mddev;
3076 newpoolinfo->raid_disks = raid_disks * 2;
3077
3078 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
3079 r1bio_pool_free, newpoolinfo);
3080 if (!newpool) {
3081 kfree(newpoolinfo);
3082 return -ENOMEM;
3083 }
3084 newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
3085 GFP_KERNEL);
3086 if (!newmirrors) {
3087 kfree(newpoolinfo);
3088 mempool_destroy(newpool);
3089 return -ENOMEM;
3090 }
3091
3092 freeze_array(conf, 0);
3093
3094
3095 oldpool = conf->r1bio_pool;
3096 conf->r1bio_pool = newpool;
3097
3098 for (d = d2 = 0; d < conf->raid_disks; d++) {
3099 struct md_rdev *rdev = conf->mirrors[d].rdev;
3100 if (rdev && rdev->raid_disk != d2) {
3101 sysfs_unlink_rdev(mddev, rdev);
3102 rdev->raid_disk = d2;
3103 sysfs_unlink_rdev(mddev, rdev);
3104 if (sysfs_link_rdev(mddev, rdev))
3105 printk(KERN_WARNING
3106 "md/raid1:%s: cannot register rd%d\n",
3107 mdname(mddev), rdev->raid_disk);
3108 }
3109 if (rdev)
3110 newmirrors[d2++].rdev = rdev;
3111 }
3112 kfree(conf->mirrors);
3113 conf->mirrors = newmirrors;
3114 kfree(conf->poolinfo);
3115 conf->poolinfo = newpoolinfo;
3116
3117 spin_lock_irqsave(&conf->device_lock, flags);
3118 mddev->degraded += (raid_disks - conf->raid_disks);
3119 spin_unlock_irqrestore(&conf->device_lock, flags);
3120 conf->raid_disks = mddev->raid_disks = raid_disks;
3121 mddev->delta_disks = 0;
3122
3123 unfreeze_array(conf);
3124
3125 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3126 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3127 md_wakeup_thread(mddev->thread);
3128
3129 mempool_destroy(oldpool);
3130 return 0;
3131}
3132
3133static void raid1_quiesce(struct mddev *mddev, int state)
3134{
3135 struct r1conf *conf = mddev->private;
3136
3137 switch(state) {
3138 case 2:
3139 wake_up(&conf->wait_barrier);
3140 break;
3141 case 1:
3142 freeze_array(conf, 0);
3143 break;
3144 case 0:
3145 unfreeze_array(conf);
3146 break;
3147 }
3148}
3149
3150static void *raid1_takeover(struct mddev *mddev)
3151{
3152
3153
3154
3155 if (mddev->level == 5 && mddev->raid_disks == 2) {
3156 struct r1conf *conf;
3157 mddev->new_level = 1;
3158 mddev->new_layout = 0;
3159 mddev->new_chunk_sectors = 0;
3160 conf = setup_conf(mddev);
3161 if (!IS_ERR(conf))
3162
3163 conf->array_frozen = 1;
3164 return conf;
3165 }
3166 return ERR_PTR(-EINVAL);
3167}
3168
3169static struct md_personality raid1_personality =
3170{
3171 .name = "raid1",
3172 .level = 1,
3173 .owner = THIS_MODULE,
3174 .make_request = raid1_make_request,
3175 .run = raid1_run,
3176 .free = raid1_free,
3177 .status = raid1_status,
3178 .error_handler = raid1_error,
3179 .hot_add_disk = raid1_add_disk,
3180 .hot_remove_disk= raid1_remove_disk,
3181 .spare_active = raid1_spare_active,
3182 .sync_request = raid1_sync_request,
3183 .resize = raid1_resize,
3184 .size = raid1_size,
3185 .check_reshape = raid1_reshape,
3186 .quiesce = raid1_quiesce,
3187 .takeover = raid1_takeover,
3188 .congested = raid1_congested,
3189};
3190
3191static int __init raid_init(void)
3192{
3193 return register_md_personality(&raid1_personality);
3194}
3195
3196static void raid_exit(void)
3197{
3198 unregister_md_personality(&raid1_personality);
3199}
3200
3201module_init(raid_init);
3202module_exit(raid_exit);
3203MODULE_LICENSE("GPL");
3204MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
3205MODULE_ALIAS("md-personality-3");
3206MODULE_ALIAS("md-raid1");
3207MODULE_ALIAS("md-level-1");
3208
3209module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);
3210